See the test data.
ISE happened because at some point after incomplete `getValue` resolution
of `a` property, we updated in the `transformAccessors` the property type
to the `Variable(Y)` type and then used it as a 3rd argument for
`setValue` call which is incorrect because the variable belongs
to a different constraint system (from `getValue`).
Mostly, the fix is just a repeating K1 behavior, namely postponing
`setValue` resolution until delegate inference is completed.
^KT-59066 Fixed
Denis.Zharkov